Ambient seismic noise monitoring of a clay landslide : To- To-ward failure prediction 1To-ward failure prediction1

Dans le document The DART-Europe E-theses Portal (Page 187-200)

Surveillance sismique passive

7.2 Ambient seismic noise monitoring of a clay landslide : To- To-ward failure prediction 1To-ward failure prediction1


Les glissements de terrain argileux peuvent connaitre des ´episodes catastrophiques d’acc´e-l´eration (coul´ees de terre, coul´ees de boue) qui posent de s´erieux probl`emes pour la gestion de l’al´ea `a travers le monde. Le m´ecanisme propos´e d’´ecoulement conduisant `a ces types de mouve-ments est le plus souvent l’augmentation de la pression interstitielle dans la masse, g´en´erant une liqu´efaction partielle ou compl`ete de celle-ci. Il en r´esulte une transition solide-liquide par une r´e-duction spectaculaire de la rigidit´e m´ecanique dans les zones liqu´efi´ees, qui pourrait ˆetre d´etect´ee par la surveillance des variations de vitesse d’onde de cisaillement. La technique de corr´elation du bruit sismique ambiant a ´et´e appliqu´ee pour mesurer la variation de la vitesse de l’onde sismique de surface dans le glissement de terrain du Pont Bourquin (Alpes suisses). Ce petit mais actif glissement-coul´ee a ´et´e ´equip´e de capteurs passifs dans le but d’enregistrer en continu le fond diffus de bruit ambiant au cours du printemps et de l’´et´e 2010. Un glissement-coul´ee de quelques milliers de m`etres cubes est survenu `a la mi-Aoˆut 2010, apr`es une intense p´eriode pluvieuse. Cet article montre que la vitesse sismique du mat´eriau de glissement, mesur´ee `a partir de corr´elo-grammes quotidiens du bruit ambiant, diminue continuellement et rapidement pendant plusieurs jours avant l’´ev´enement catastrophique. `A partir d’une analyse spectrale de la diminution de la vitesse, il a ´et´e possible de d´eterminer l’emplacement du changement rh´eologique `a la base de la couche de glissement. Ces r´esultats d´emontrent que le bruit ambiant sismique peut ˆetre utilis´e pour d´etecter les variations de rigidit´e avant le d´epart d’un tel glissement et pourrait ˆetre utilis´ee pour les pr´edire.


Given that clay-rich landslides may become mobilized, leading to rapid mass movements (earthflows and debris flows), they pose critical problems in risk management worldwide. The most widely proposed mechanism leading to such flow-like movements is the increase in water pore pressure in the sliding mass, generating partial or complete liquefaction. This solid-to-liquid transition results in a dramatic reduction of mechanical rigidity in the liquefied zones, which could be detected by monitoring shear wave velocity variations. With this purpose in mind, the ambient seismic noise correlation technique has been applied to measure the variation in the seismic surface wave velocity in the Pont Bourquin landslide (Swiss Alps). This small but active composite earthslide-earthflow was equipped with continuously recording seismic sensors during spring and summer 2010. An earthslide of a few thousand cubic meters was triggered in mid-August 2010, after a rainy period. This article shows that the seismic velocity of the sliding material, measured from daily noise correlograms, decreased continuously and rapidly for several days prior to the catastrophic event. From a spectral analysis of the velocity decrease, it was possible to determine the location of the change at the base of the sliding layer. These results

1. Auteurs : G. Mainsant (ISTerre), E. Larose (ISTerre), C. Br¨onnimann (GEOLEP), D. Jongmans (ISTerre), C. Michoud (IGAR, UNIL), M. Jaboyedoff (IGAR, UNIL). Publi´e dans Journal of Geophysical Research le 22 mars 2012.

7.2. Ambient seismic noise monitoring of a clay landslide : Toward failure prediction 173

demonstrate that ambient seismic noise can be used to detect rigidity variations before failure and could potentially be used to predict landslides.

7.2.1 Introduction

All mountainous areas are affected by gravitational mass movements of various types, sizes and velocities, which could have a major impact on life and property. Landslides in clay-rich formations, which are widespread over the world, are characterized by unpredictable accelera-tion and liquefacaccelera-tion phases (Iverson et al.(1997) ; Malet et al. (2005)). Of particular concern for hazard assessment is the triggering of earthflows and debris flows, the rheology of which switches from solid to fluid. This phenomenon has been widely reported in all types of recent clay deposits, including Quaternary marine sensitive (Crawford (1968) ;Eilertsen et al.(2008)) or nonsensitive clays (Picarelli et al.(2005)) and lacustrine clay deposits (Bi`evre et al.(2011a)).

But flow-like movements have also been frequently observed in fractured and weathered clay-rich rocks, such as shales, marls and flyschs (Angeli et al.(2000) ;Picarelli et al.(2005) ;Malet et al.

(2005)), and in volcanic rocks in which primary minerals were altered to clays (Coe et al.(2003)).

Predicting these sudden events, primarily controlled by groundwater conditions, has been an active research topic for the last two decades (Lee and Ho (2009)). Empirical prediction me-thods have been proposed, relying on two types of measurements : (1) surface displacements, whose change to rupture is usually interpreted using slope creep theories (Petley et al. (2005)), and (2) hydrological factors such as precipitation, soil water content or pore pressure, used as predictors with threshold values determined in an empirical or statistical manner (Keefer et al.

(1987)). Although these empirical methods have been successfully applied in some specific cases, they do not provide a real insight into the mechanisms involved, and have proved to be very sen-sitive to changes in landslide geometry and deformation. Recently, theoretical models coupling a slope instability mechanism and hydrological modeling have been developed for predicting landslide occurrence (Keefer et al. (1987) ; Crosta and Frattini (2008)). However, in 3D, such approaches require considerable investigation and computational effort.

For debris flows and earthflows occurring in fine-grained soils during or after heavy and sustained rainfalls, the triggering mechanism most often proposed is the partial or total liquefaction of the mass, resulting from an increase in pore water pressure (Picarelli et al.(2005) ;Van Asch et al.

(2007)). As the shear wave velocity (Vs) in a fluid tends to 0 (Reynolds (1997)), the bulk shear wave velocity should dramatically decrease in the vicinity of liquefied zones. Moreover, it has been recently shown that, in a clay-rich landslide,Vs also significantly decreases with the extent of damage in the material (Renalier et al.(2010)). This suggests that continuousVsmeasurement could be valuable for monitoring clay slope degradation and would constitute an alternative to the classical prediction methods.Vsis usually obtained from active source-receiver experiments.

However, the reproducibility of seismic sources is very limited, and it is difficult to ascertain whether seismic response changes actually result from a change in the mechanical properties of the medium or from the source. The ambient noise correlation technique developed over the last 10 years (Weaver and Lobkis (2001) ;Shapiro et al.(2004)) offers a realistic alternative to using controlled sources. The local Green’s function (or impulse response) can in fact be determined

from the cross correlation of ambient noise continuously acquired by two passive sensors as if one of them was a source. This method has found considerable applications in seismic imaging at different scales (Shapiro et al., 2005 ; Larose et al., 2006). More recently, it was demonstrated that the tail portion of the correlograms, the so-called coda part formed by scattered waves, is extremely sensitive to small changes in the medium (Sens-Sch¨onfelder and Wegler (2006) ; Brenguier et al. (2008a), Brenguier et al. (2008b)). By comparing the phases of the waves in the coda, apparent relative velocity changes of the material can be measured with a precision better than 0.1%. This can be performed even if the correlograms do not give the exact Green’s function between the sensors. Correlograms are however required to be stable in time, implying a relatively constant background noise over the period of interest (Hadziioannou et al. (2009)).

In the present manuscript we will apply the noise correlation technique on a landslide where the noise is in part due to traffic on the road, which constitutes a spatially stable background noise. The purpose of the paper is to detect mechanical changes in an active clay landslide where failure is expected.

7.2.2 The Pont Bourquin Landslide History and Geology Historical Context

The Pont Bourquin landslide is located in the Swiss Prealps, 40 km to the east of the town of Lausanne (Figure 7.2). Although the whole area has been affected by landslide phenomena since the last glacial retreat (more than 10,000 years ago), aerial photos show that gravitational deformation appeared in the mid 90s in the upper part of the hillside and that the slope insta-bility gradually developed over a period of about 10 years (Jaboyedoff et al. (2009)). In 2006, displacements of up to 80 cm created the head scarp of a 240 m long translational landslide affecting an area of about 8,000 m2, with a width varying from 15 m to 60 m (Figure 7.2). On 5 July 2007, a 3 day period of heavy rainfall (cumulative depth of 95 mm) triggered an earthflow, which started from the main secondary scarp (SS in Figures 1 and 2) and cut the frequently used Pillon Pass road located at the toe of the Pont Bourquin landslide. This earthflow, with an estimated volume of 3,000 to 6,000m3, affected a layer a few meters thick in the transporta-tion area (TA) of the Pont Bourquin landslide (Jaboyedoff et al.(2009)). During the following 3 years, the entire landslide has exhibited a general translational motion associated with high internal deformation and numerous small superficial translational or rotational landslides, ear-thflows and debris flows. These multiple erosive processes gradually created a bulge of highly deformed material (accumulation zone labeled AZ in Figures 1 and 2) that progressively loaded the lower part of slope (see also Text S1 in the auxiliary material). This material accumulation led to the toe failure between 18 and 20 August 2010, following significant cumulative rainfall in July. Geological Context

According to the geological map (Badoux et al. (1990)) the Pont Bourquin landslide is loca-ted in a tectonically very complex zone. Three thrust faults dipping approximately 35˚ toward the North cross the landslide and separate distinctive geological formations (Figure 7.3a). In the upper and lower parts of the slope, the bedrock is composed of Triassic cargneule (cellular

7.2. Ambient seismic noise monitoring of a clay landslide : Toward failure prediction 175





E2 S1 S2






6°E 8°E 10°E

46°N 47°N



Figure7.2 – Aerial photo of the Pont Bourquin landslide in June 2009, with the location of the two electrical profiles E1 and E2, the two seismometers S1 and S2 installed on sta-ble ground, and the inclinometer I1. The headscarp (HS), main secondary scarp (SS), transportation area (TA), and accumulation zone (AZ) are also indicated. The Pont Bour-quin landslide (red cross) is located on the topographic inset map of Switzerland (L, Lausanne ; Z, Zurich). The gravita-tional instability threatens the road carrying heavy traffic over the Pillon pass (bottom of the photo).

dolomite) associated with gypsum. These highly soluble and deformable rocks could have pro-moted slope destabilization at the landslide toe. Below the cargneule layer, the upper part of the slope is made of Aalenian black shale, the weathering of which is the main source of the sliding clay material. In the middle part of the slope, the landslide overlies flysch consisting of thin-bedded turbidites including siltstone and conglomerate. The top of the hill is covered by several meters of moraine deposits. The rocks have been heavily fractured by the Alpine orogeny and subsequently affected by toppling, chemical weathering and freeze and thaw cycles, which contributed to a high degree of fragmentation of the outcropping rocks. These alterations have resulted in muddy material that can give rise to numerous small earthflows and debris flows along the slope. Deposits resulting from ancient mass movements locally cover the lower part of the slope. The present day landslide mass is mainly composed of a mixture of moraine material, mainly visible in the upper part, and weathered debris from the Aalenian black shale, flysch sandstone and marl alternations, making the sliding material predominantly clayey. According to the classification proposed by Cruden and Varnes (1996), this landslide can be termed an active composite earthslide-earthflow.

Figure 7.3 – (a) North-south geological cross section along the Pont Bourquin landslide, constructed from the local geological atlas (Badoux et al.(1990)) and the electrical images shown in Figure 2b. (b) North-south and east-west oriented electrical resistivity tomography profiles (see location in Figure 7.2). The head-scarp (HS), main secondary scarp (SS), transportation area (TA), and accumulation zone (AZ) are also indicated.

7.2. Ambient seismic noise monitoring of a clay landslide : Toward failure prediction 177 Geophysical Investigation

In order to clarify the landslide geometry and the geological structure underneath, two electrical resistivity tomography (ERT) profiles E1 and E2 were collected (see location in Fi-gure 7.2), along and perpendicular to the slope, respectively. Data were acquired using the Wenner-Schlumberger configuration (Dahlin and Zhou (2004)) with 64 electrodes and an elec-trode spacing of 5 m and 1.5 m for E1 and E2, respectively. Data were inverted through a least squares inversion (L2-norm) using the RES2DINV software (Loke, 1998). ERT images have been obtained for a RMS value lower than 5%. Electrical images are shown in Figure 7.3b. The su-perficial clay-rich sliding layer is clearly evidenced by a resistivity lower than 100 ohm m, with a thickness varying from a few meters to locally more than 10 m along the profile. This low resistivity results from the high percentage of saturated clay in the sliding mass and from the high salinity of the water (total salinity greater than 1500 mg/l in superficial water between S1 and S2, Figure 7.2). The potentially mobilized volume of the whole landslide is estimated to be 30,000 to 40,000 m3. Below the sliding material, the cargneule and gypsum formations at the top and bottom of the slope can be distinguished by their higher resistivity (from 200 to 500 ohm m in the cargneule and up to 2000 ohm m in gypsum). Conversely, the black shales are characterized by low resistivity values ranging from 100 to 200 ohm m. Finally, the flysch formation has a resistivity between 200 and 500 ohm m, a range similar to that measured for the cargneule. The combination of the two electrical images and geological observations has yielded the interpretative cross section of Figure 7.3a.

Two active seismic profiles were performed along and across the landside (same location as the electrical profiles E1 and E2). The surface wave inversion technique was applied to 8 signals recorded in the accumulation zone of the landslide to infer the shear wave velocity profile in this area. For the longitudinal profile, signals were generated with explosive shots and recorded by 8 geophones 5 m apart (channels 21 to 28, Figure 7.4a). For the second transverse profile (between S1 and S2), the source was a hammer striking a plate, and the records from 8 geophones located within the landslide (4 m intertrace distance) were processed. The Rayleigh wave phase velocity dispersion curves were computed along the two perpendicular travel paths (Figure 7.4b), using the frequency-wave number technique (Lacoss et al., 1969). The two dispersion curves cover the 10−30 Hz frequency range, according to the frequency of the sources (explosive and hammer), and partly overlap around 15 Hz. The 10% difference in phase velocity around 15 Hz (500 to 550 m/s) probably results from different spatial variations along the two profiles. Dispersion curves were inverted using the enhanced neighborhood algorithm (Wathelet (2008)), assuming a 1D structure along the two directions below the accumulation zone. The misfit function is defined by the following equation :

where cdi is the phase velocity of the data curve at frequency fi, cci is the velocity of the calculated curve at frequency f, and n is the number of frequency samples. The inversion was constrained by imposing a thickness higher than 10 m for the clay-rich sliding layer, in agreement

Figure 7.4 – Shear wave velocity determination from the Rayleigh waves measured along two perpendicu-lar profiles (same location as E1 and E2, Figure 7.2). (a) Normalized raw signals along profile 1. The spacing between geophones is 5 m. (b) Phase velocity dispersion curves computed from profiles 1 (triangles) and 2 (circles). (c) Vs profiles resulting from the inversion of dispersion curves with a three-layer model. (d) Dispersion curves corresponding to the models shown in Figure 7.4c.

7.2. Ambient seismic noise monitoring of a clay landslide : Toward failure prediction 179

with the electrical data. Figure 7.4c shows the computed S wave velocity profiles with the misfit values for a three-layer model. The shear wave velocity in the superficial layer of a few meters thick is poorly constrained, owing to the lack of information at high frequency. Below, the best fitting models (misfit lower than 5%) show that the seismic velocity in the landslide is between 360 and 420 m/s. The bottom of this layer is found at a depth of about 11 m. Below this depth, the velocity increased to about 640 m/s in the undisturbed layers. Figure 7.4d displays all dispersion curves corresponding to the models obtained, with good agreement being shown between models and observations. Groundwater Level Monitoring

The level of the water table was measured in one 5 m deep borehole located in the accumu-lation zone (see Figure 7.5 for location). The system consists of a piezometer sensor connected to a data logger operated continuously, and a barometer to correct the water table height from the atmospheric pressure fluctuations. During the experiment time (April to August 2010), the measured water table fluctuated from 3.7 m depth to less than 1 m (see section 4.1).

7.2.3 Displacement Monitoring Surface Displacement From GNSS and Electronic Distance Meter

In order to evaluate the activity of the Pont Bourquin landslide, the displacement of twelve targets placed on the surface was recorded. Three targets (I1, T10 and T11, see location in Figures 1 and 4), were monitored from summer 2009 until August 2010 using a Differential Global Navigation Satellite System (D−GNSS) (of Engineers (2003)). The data were acquired by two Topcon HiPerRPro antennas tracking their position from the Russian and American geodesic satellite constellations. The field procedure followed the Real Time Kinematic (RTK) method.

A base station antenna was set up on a unique reference point location close to the landslide (about 650 m away). Targets on the landslide were 3D located with the second GNSS antenna (rover station), using the correction information communicated by the base station. Instrumen-tal accuracy is ∼12 mm [Gili et al. (2000)], which is considered negligible with regards to the observed meter-scale displacements. Nine additional targets (T1–T9) were installed in spring 2010 and have been periodically monitored with a Topcon GPT-9003M reflector total station [of Engineers (2007)]. For each measurement campaign, the device was first installed at a refe-rence point (the same as the GNSS base station) having a direct line of sight to the landslide and orientated by shooting at a reference prism located in a stable area close to the landslide, the coordinates of which were measured by D−GNSS.

The surface displacements presented in Figure 7.5 exhibit an acceleration during summer 2010, before the slope failure. Active creeping in and above the accumulation zone was evidenced by D−GNSS and EDM data (arrows in Figure 7.5). From July 2009 to May 2010 (green arrows), displacement rate values were lower than 1 m/month. From 21 May to 21 July 2010 (orange arrows), a mean displacement rate of 6 m/month was measured by four targets in the trans-portation zone. The motion in this zone accelerated from 21 July to 23 August 2010 (purple


Landslide activated in August 2010 Slided mass Spring

[ 07.21.2009 - 05.20.2010 ] [ 07.21.2010 - 08.23.2010 ] [ 05.21.2010 - 07.21.2010 ] [ 05.21.2010 - 08.23.2010 ]


Figure 7.5 – Mean velocity (meters per month) of 12 targets (T1T11 and I1), monitored since July 2009 by Differential Global Navigation Satellite System (DGNSS) and May 2010 by Electronic Distance Meter (EDM). In the top part of the landslide, displacements did not exceed 1 m between 20 May 2010 and 23 August 2010, while they exceeded 20 m in the middle of the transportation area during these 3 months, which considerably loaded the accumulation zone (AZ). (Topography outside the landslide : highresolution Digital Elevation Model data from Swisstopo ; topography inside the landslide : terrestrial laser scanning (TLS) data from UNIL.)

7.2. Ambient seismic noise monitoring of a clay landslide : Toward failure prediction 181









-2.00 Differences

(in m) along the

Y axis

S1 S2



~10 m


I2 G

Figure 7.6 – Differences (in m, southward, i.e., along the y axis) between two point clouds acquired by TLS on 19 May 2010 and on 20 July 2010. Positive (accumulated material) and negative (eroded material) movements are shown in red and blue, respectively. Black indicates no data, and gray indicates unreliable data. Red dashed lines isolate particular areas of the landslide. Small erosion of the head scarp (HS) by flowing processes. Retrogression of the most active secondary scarp (SS) through small and discontinuous translational landslides. Very active creeping inside the transportation area (TA). The locations of the inclinometer (I1) and the seismometers (S1 and S2) are indicated. The accumulation zone located between S1 and S2 could not be monitored by TLS because of forest cover.

arrows), when displacements ranging from 17 m up to 21 m were recorded in one month on the same targets. In the meantime, sliding velocities at the head and the secondary scarps were lower than 0.5 m/month. Terrestrial Laser Scanning

Terrestrial laser scanning (TLS) was regularly performed from April to September 2010 in order to monitor ground motions over the whole slope. TLS is a remote sensing technique capable of obtaining local images of the earth’s 3D topography by acquiring point clouds of the ground position (Baltsavias (1999) ;Lichti et al.(2002)). TLS involves sending a laser pulse in a known direction and the distance is evaluated by measuring the return time of the pulse reflected by the ground surface. Scanning on a regular grid provides images of several million points. The TLS device was an Optech ILRIS−3D–ER using a laser with a 1500 nm wavelength and with maximum acquisition distance ranging from 800 to 1200 m.

The TLS data acquisition and processing followed these main stages : (1) the point clouds of the upper part of the Pont Bourquin landslide were acquired from the same scanning point of view at different epochs with a mean resolution of 30 mm (average distance between points) ; (2) two

The TLS data acquisition and processing followed these main stages : (1) the point clouds of the upper part of the Pont Bourquin landslide were acquired from the same scanning point of view at different epochs with a mean resolution of 30 mm (average distance between points) ; (2) two

Dans le document The DART-Europe E-theses Portal (Page 187-200)